At Microsoft Croatia, I led the architecture of a cloud-hosted contextual content service, connecting user-facing APIs, session management, AI integration, and asynchronous processing. I chose hybrid retrieval in Azure AI Search, combining vector and keyword matching with semantic ranking; on a reviewed question set, this cut unsupported answers from 14% to 8%.
I engineered the Python RAG pipeline with LangChain and Azure OpenAI, and separated session state from AI context to reduce average prompt size by roughly 35%. Streaming answers through Server-Sent Events also cut time to first visible text from about three seconds to under one.
Previously at Microsoft, I designed shared backend services and messaging, calendar, and synchronization workflows used across applications. I’ve worked across Python and Node.js services, cloud infrastructure, and data pipelines, and guided engineers through architecture and code reviews.

